Big general change in the API where basically all the {application}_data_from_{thing} disappeared to give birth to:
ImageDataBunchTextDataBunchTabularDataBunch
This is motivated by the fact each of those are going to have helper functions specific to their application (ImageDataBunch is gaining a lot of them but more will follow) and it also simplifies a part that was a bit obscure in NLP, the data_func=something arguments. Now we subclass TextDataBunch according to the task in TextLMDataBunch or TextClasDataBunch.
Docs and examples are all update to reflect that.